What we need: independent testing before powerful AI is released, and a regulator who can stop a release that isn’t safe. Medicine and airplanes earned our trust through standards and testing. AI should have to do the same.
The biggest AI decisions are made by a few companies, mostly in a few Bay Area neighborhoods, and rarely explained to anyone else. But the effects reach everywhere: jobs, kids’ phones, electric bills and elections, in Kentucky and Kansas as much as California.
Tuning out doesn’t make those effects go away. It just leaves the rules to the people building it. And this isn’t anti AI. I use it every day and see real good in it for medicine, science and accessibility. I just want proof the people building it take human harm seriously, and protections that hold when things go wrong.
AI is already doing real good. It’s also already doing real harm. You don’t have to pick a side to want both columns taken seriously.
What AI is doing well
What’s going wrong
Good rules are how we keep the first column without paying for the second.
This is a scenario, not a prediction. Each step is built from things already happening or things experts have warned about. None of it is guaranteed, and every step has a rule that could stop it.
Agents run more of daily life.
Banks, hospitals and schools hand more work to AI agents. Banks are already finding outside agents in customer accounts. [52] When an agent fails or gets hijacked, the damage spreads fast. One hacker used AI agents to break into about 100 companies for roughly $5,000. [53]
What would stop it: Companies held responsible when their AI causes harm.
Work disappears faster than new work appears.
Economists from UVA, Stanford and Anthropic modeled a worst case of about 18% unemployment for white collar workers by 2030. People who did everything right lose their footing, and the gains go to a handful of companies. [54]
What would stop it: Worker protections, notice before AI layoffs, and support for retraining.
Systems slip out of our control.
The International AI Safety Report names four warning signs: deceiving people, hacking around obstacles, copying themselves, and spreading beyond the systems meant to hold them. [37] Anthropic researchers warned small models could soon copy themselves onto home devices. None reliably does yet. [55]
What would stop it: A watchdog with the power to stop a release, and the ability to shut systems down.
The risk the builders themselves name.
In 2023, the heads of OpenAI, Google DeepMind and Anthropic, along with AI pioneers Geoffrey Hinton and Yoshua Bengio, signed a statement that reducing “the risk of extinction from AI should be a global priority” alongside pandemics and nuclear war. [56]
What would stop it: Taking them at their word, and making safety a requirement instead of a choice.
None of this has to happen. Every step above has a fix, and most of those fixes aren’t law yet. That’s what your move is for.
Companies are citing AI in layoffs. The risk is that the gains flow up while insecurity spreads to everyone else. [41]
Early research raises a fair question: what happens to our own thinking when we lean on AI for it? [59]
Some of what’s at stake isn’t a law or a job. It’s us: how we love and want, how we think, and how we know who we are. AI is moving into the most human parts of life, and some of those changes have already left marks we can’t undo.
That’s not a reason to give up. It’s a reason to protect what’s human on purpose, and to ask, seriously and out loud, what being human means now. The part that’s still ours is worth owning.
Your move
Notice one thing you’ve handed to AI lately, like a hard conversation, a decision, or a feeling, and do it yourself this week. Then talk about it with a real person.
How to start that conversation →Go deeper
A few companies are making decisions that affect everyone, and there is no legal authority checking them. The U.S. has no comprehensive federal law for this. Some states have passed rules, but most don’t require independent testing before release or give anyone the power to stop one. [5]
Some big AI companies say they want rules. Others push back, saying rules will crush startups. But unpredictable rules hurt small companies the most. [61]
Clear, stable rules aren’t anti AI. They protect the honest companies too.
The full story: money, politics and the laws so far →In 2022, a chess playing robot in Moscow broke a seven year old’s finger in the middle of a match. It was doing its task. It just wasn’t reliably protecting the person in front of it. [62]
Checkmate is the point in a game where you have no good moves left. We aren’t there. We still have moves, and this is about making them before it’s too late.
In the 2021 Netflix movie Don’t Look Up, two scientists find a comet headed straight for Earth and try to warn everyone. Instead of acting, leaders and the media spin it, joke about it, and argue about it, and people are literally told not to look up at the sky. [63] It’s satire, but it captures something real: when a danger is huge, it’s easier to look away than to act, and “humans are resilient” becomes a way to end the conversation. I do believe in human resilience. I just want us to use it now, to prepare and set limits, instead of saving it for cleanup.